Emergence, Not Supervenience

advertisement
`Emergence, Not Supervenience', Philosophy of Science 64 (1997),
pp. S337-S345
EMERGENCE, NOT SUPERVENIENCE
Paul Humphreys‡
ABSTRACT
I argue that supervenience is an inadequate device for
representing relations between different levels of phenomena. I
then provide six criteria that emergent phenomena seem to
satisfy. Using examples drawn from macroscopic physics, I
suggest that such emergent features may well be quite common in
the physical realm.
1. Introduction
Supervenience has for years been the tool of choice for
anti-reductionists. For reasons that are now familiar -- the
inability to derive theories about a higher level domain from
theories about a more fundamental domain, either because of
syntactic complexities or the lack of the appropriate concepts
in the lower domain; the multiple instantiability of concepts in
the mental realm, such as `pain', and the resulting
implausibility of reducing these concepts to unitary concepts in
a lower realm; the mismatch of natural kind terms within laws at
different levels, and so on -- reduction has been out and
1
supervenience has been in. This position is only half right.
Reduction is still not an option, but supervenience is no good
either. It is a notion that
______________________________________________________________
‡Corcoran Department of Philosophy, 521 Cabell Hall, University
of Virginia, Charlottesville, VA 22903, USA. pwh2a@virginia.edu
is empty of any scientific content, and what antireductionistsneed in its place is emergence. The latter idea can
properly capture the picture of distinctively different layers
of the world in which anti-reductionists believe.
As always,
there is a price to pay for adopting an emergentist view. In
certain cases one must give up a number of metaphysical views
that, for good or ill, have proven attractive to many. One of
these is ontological minimalism.
Ontological minimalism runs roughly like this: a) There is
a relatively small set of fundamental constituents of the world,
b) to individuate these we need only intrinsic (i.e. nonrelational) properties, and c) all the non-fundamental
individuals and properties are composed of or from these
fundamental entities. Ontological minimalism is an attractive
view. Indeed, if one accepts a well-known set of philosophical
doctrines, it comes out almost right. Chief amongst these
2
doctrines are 1) Giving primacy to logical reconstructions of
ordinary and scientific concepts, and 2) a broadly Humean
account of causal relations. If, in contrast, one thinks that
these two doctrines (and related ideas) are wrong, then
ontological minimalism will seem to be a perverse
misrepresentation of the way our world happens to be. This is
not the place to argue in detail against those two doctrines
(arguments for their rejection can be found in my (1995) and
(1997a)1). Suppose that one did accept the rough picture provided
by these two doctrines (which I do not). Then the core belief
that underlies ontological minimalism is that non-fundamental
entities are nothing but collections of fundamental entities.
What is important for present purposes is that ontological
minimalists commonly interpret the `nothing but' as either
`aggregate' or `supervenient upon'.2
2. Supervenience
In his paper "Emergence as Non-Aggregativity" William
Wimsatt (1998) has demonstrated beyond any doubt that
aggregativity is a hopelessly inadequate approach to almost any
composite object to which it has been applied. Here I shall try
to show that supervenience is similarly inadequate, albeit in a
different way.
Despite the fact that many supervenience accounts are non-
3
reductive in intent, there is a strong residue of suspiciously
reductionist terminology within many of them. There is the `If A
supervenes upon B, then A is nothing but B,' talk of course. In
addition, there is the idea that if A supervenes upon B, then
because A's existence is necessitated by B's existence, all that
we need in terms of ontology is B.3 This is a core belief that
motivates physicalism -- all one needs is the ontology of
physical objects and properties, then everything else supervenes
upon those.
From the plethora of definitions of supervenience that have
been offered I shall take as representative that of strong
supervenience, which seems to be the most widely accepted. My
criticisms will apply, mutatis mutandis, to weak and global
supervenience.
One standard definition of strong supervenience is this:
Definition: A family of properties M strongly supervenes on a
family N of properties iff, necessarily, for each x and each
property F in M, if F(x) then there is a property G in N such
that G(x) and necessarily if any y has G it has F. (Kim 1993,
65)
Such definitions have little content until the kind of
necessitation involved has been given. So let us consider
nomological necessity and logical (or metaphysical) necessity
4
(for both operators) as natural interpretations of the modal
operators.4
Supervenience, as its name implies, is standardly taken to
be a relation between two collections of properties, one at a
higher level than the other. Yet there is nothing in the
definition itself that requires the relation to hold only
between different levels, and this raises an interesting
question: why is supervenience taken to be revealing as an
inter-level relation but has no significant applications as an
intra-level relation? Or, to put it another way, in what way, if
at all, does the existence of a supervenience relation between
two sets of properties explain why the members of the
supervening set have an inferior ontological status than do the
members of the subvening set? (And have no doubt that they carry
an inferior status, for how else could physicalism have the
appeal that it does if physical states were not the only
legitimate ontology?) To see why this is a problem, let us, as
it were, turn the supervenience relation on its side and apply
it within a given level. First, consider the causal relation.
Let M be a set of `effect properties', and N be a set of `causal
properties' i.e., M consists in properties, instances of which
are characteristically found as effects of instances of
properties in N. For a simple example, take M as expansion in
5
volume, and N as properties that cause expansion, such as heat,
increase in pressure, and so on. Then, with the first
interpretation of the necessities in the definition as
nomological necessities, it is easily seen that in every
possible world with the same laws as ours, if something expands,
then there is some other property such as heat that applies to
that thing and this nomologically necessitates its expansion.
Now, no one will agree that expansion is `nothing but'
application of heat. Why not? One plausible explanation is that
both heat and expansion are already considered to be respectable
physical properties, and so there is no need to bring in
supervenience to (putting it slightly pejoratively) `explain
away' the supervenient properties. A second, but quite
different, interpretation of supervenience, to the effect that
the supervening properties are real, but have some secondary
sort of status that is necessarily dependent upon their
subvenient properties5, does not seem very promising either,
because both heat and expansion are ordinarily taken to be
equally real.
Caveat: supervenience is concerned with the dependence of
one set of properties upon another. In the expansion case, it is
indisputable that an instance of expansion is nomologically
dependent upon (an instance of) application of heat but this is
6
not the issue under discussion here; it is whether the
properties themselves are so dependent. Of course, one already
knows what the difference is between the inter-level and the
intra-level applications of supervenience. In the case of interlevel applications, the inferior status of the supervening
properties is nothing more than their initially having been
considered to be less respectable than their subvenient bases,
which is why the supervenience relation was constructed in the
first place. Yet this clearly reveals that whatever the
supervenience relation does, it does not, in itself, answer our
question.That is, there is nothing in the nature of the
supervenience relation itself that will explain why 'vertical'
uses are appropriate and 'horizontal' uses are not.
Take a second case, one involving logical or metaphysical
necessitation. Here let M be the set consisting only of the
property `is triangular' and N the set consisting only of the
properties `closed', `three-sided', `Euclidean' and `polygon'.
Then it is again obvious that M supervenes upon N. To say that M
is nothing but N will not concern you if you subscribe to the
position that metaphysically equivalent properties are
identical. Although some philosophers have accepted this, many
have found it to be quite implausible, and those of us who
reject it in doing so also reject the idea that the property of
7
being triangular has some sort of secondary, dependent, status.
It thus seems that there is something wrong with applying
supervenience relations within ontological levels, and the
reason why is pretty clear. In both of the examples I have
described, we have good reason to hold that the supervenient
property has some independent standing, irrespective of the fact
that there are necessitation relations holding between it and
other properties. With inter-level applications of
supervenience, I doubt that anyone holds that the existence of a
supervenience relation is itself a reason for ascribing some
sort of secondary status to the supervenient property -- this is
already assumed. Mental properties, aesthetic properties, moral
properties are antecedently taken to be less desirable than
physical properties. So the question that remains is this: what
is it that makes the supervenience relation admissible as an
inter-level relation, but renders it wrong-headed as an intralevel relation? It cannot be simply a prejudice in favor of
physical properties, for that would not explain why it gives
unacceptable answers for geometrical properties. I believe that
the answer is this: supervenience is acceptable as a consistency
condition on the attribution of concepts, in that if A
supervenes upon B, you cannot attribute B to an individual and
withhold A from it. If aesthetic merit supervenes upon just
8
spatial arrangements of color on a surface, and you attribute
beauty to the Mona Lisa, you cannot withhold that aesthetic
judgement from a perfect forgery of the Leonardo painting. But
supervenience does not provide any understanding of ontological
relationships holding between levels. For that emergence is
required.
3. Possible Criteria for Emergence
Perhaps the most characteristic feature of emergent
properties is that they are novel. Novelty can mean many things,
but the simplest and crudest interpretation is that a previously
uninstantiated property comes to have an instance.6 Clearly, it
is important how this novel instance came about. If the new
value of, say, mass comes about by mere rearrangement of
existing matter (by aggregating bits of smaller mass, or slicing
up a larger mass) then we say that the 'novel' property instance
was there all along, it just was not instanced by the object
that now has it. (In the aggregate case, it was a collection of
spatially separated objects that had the 'new' mass; in the
slicing case, then it was a part of the object that had it). The
novelty criterion for emergence would then be violated, and so
this turns out to be no case of emergence at all. Suppose in
contrast that the new mass value occurred spontaneously (from
nothing), or was the result of (spontaneous) conversion of
9
energy to mass. Here there is more of a sense of novelty,
emergence from something else or from nothing. But this is a
boring kind of emergence, and not the kind of case in which we
are interested, for what we want from emergence is not a new
value of an existing property, but a novel kind of property.
This then gives us the second characteristic feature of emergent
properties, which is that they are qualitatively different from
the properties from which they emerge.
Four other features of emergent properties suggest
themselves. The third is that an emergent property is one that
could not be possessed at a lower level -- it is logically or
nomologically impossible for this to occur. The fourth is that
different laws apply to emergent features than to the features
from which they emerge. This anti-reductionist view has been
suggested by the physicist P.W. Anderson: "Reductionism is the
thesis that all of the animate and inanimate matter of which we
have any detailed knowledge is assumed to be controlled by the
same set of fundamental laws..." (Anderson 1972, 393)) We should
then have that entities of type B are emergent from entities of
type A iff entities of type B have type A entities as
constituents and there is at least one law that applies to type
B entities that does not apply to type A entities.
The fifth characteristic feature of emergent properties is
10
that they result from an essential interaction between their
constituent properties, an interaction that is nomologically
necessary for the existence of the emergent property. This is
one sense that gives us the novelty -- before the property
instances interacted there never had been an instance of the new
property. We might want to add that the property isntance occurs
at a higher level, or that the property instance is attached to
a new kind of entity. That is, a hydrogen atom is an emergent
entity - it is not just an electron and a proton put in a new
spatial arrangement, but there is an essential interaction
between them.
And the sixth feature is that emergent properties are holistic
in the sense of being properties of the entire system rather
than local properties of its constituents.
A non-relational property of a whole is emergent if it
supervenes upon but does not reduce to the non-relational
properties of its parts.(Teller, p.142) <tie in with internal
relations>
4. Some Examples
I do not suggest that any emergent phenomenon must satisfy
all of these criteria, for there is a wide variety of ways in
which emergence can occur. In my (1997), I provide an account of
one kind of emergence in terms of so-called `fusion' operations,
11
and it turns out that it fits the most widely canvassed examples
of emergence, those involving quantum entanglements. This sort
of emergence directly satisfies our fifth and sixth criteria
above, and when it is the basis for macroscopic phenomena such
as superconductivity and superfluidity, it will satisfy at least
criteria one, two, and four as well.7 In fact, we can look to
macroscopic phenomena for examples of other kinds of emergence
too.
The idea that emergence can involve a qualitative
difference in properties even though the object possessing those
properties is composed from more elementary objects occurs in
what is known as the theory of macroscopic systems.
A
macroscopic system can be defined as one whose equation of state
is independent of its size (Sewell 1986, 4). An ideal
macroscopic system is one containing an infinite number of
particles, the density of which is finite.
A real macroscopic
system is one that is sufficiently large that its equation of
state is empirically indistinguishable from an ideal macroscopic
system (see Sewell 1986, 4).
These definitions are purely
objective and do not rely on a characterization of `macroscopic'
by way of reference to human observers.
One of the principal
features of macroscopic systems in that sense is that they
display qualitatively different properties than do the
12
microcomponents that compose them and that they are holistic.
They thus satisfy our second and sixth criteria above. To
illustrate this, here is a typical quotation:
13
Macroscopic systems enjoy properties that are
qualitatively different from those of atoms and
molecules, despite the fact that they are composed of
the same basic constituents, namely nuclei and
electrons. For example, they exhibit phenomena such as
phase transitions, dissipative processes, and even
biological growth, that do not occur in the atomic
world. Evidently, such phenomena must be, in some
sense, collective, in that they involve the
cooperation of enormous numbers of particles: for
otherwise the properties of macroscopic systems would
essentially reduce to those of independent atoms and
molecules. (Sewell 1986, 3)
14
Also:
15
The quantum theory of macroscopic systems is
designed to provide a model relating the bulk
properties of matter to the microscopic ones of its
constituent particles. Since such a model must possess
the structure needed to accommodate a description of
collective phenomena, characteristic of macroscopic
systems only, it is evident that it must contain
concepts that are qualitatively different from those
of atomic physics. (Sewell, 1986, 3)
16
This kind of talk is reminiscent of Rohrlich's conceptual
irreducibility (Rohrlich 1997)8,
the main point being that the
emergent properties cannot be possessed by individuals at the
lower level because they occur only with infinite collections of
constituents. Some of the most important cases of macroscopic
phenomena are phase transitions, such as the transition from
liquid to solid. This transition is not exhibited by the
microcomponents of the liquid (or the solid) since the
individual components are the same in each phase.
It is their
collective relationship to each other that changes across the
(usually discontinuous) phase transition.
Thus, it is the
interactions between the constituents that makes for the
qualitatively different macroscopic behavior.
An example of
this kind of interactive behavior is the spontaneous
ferromagnetism that occurs below the Curie temperature, which is
absent above that phase transition point and which occurs
because of spin interaction coupling between adjacent particles
on the lattice. It might appear that such cases are no different
from more familiar ones, wherein the properties of water
molecules, say, are different from the properties of the
hydrogen and oxygen atoms that compose them. Essentially
macroscopic phenomena are different from these, however, as the
17
following example shows.
There are many qualitatively distinct macroscopic phenomena
associated with ferromagnets and similar systems, but one will
suffice here. If one takes a ferromagnet whose Hamiltonian is
spherically symmetric, then below the Curie temperature the
system is magnetized in a particular direction, even though
because of the spherically symmetric Hamiltonian, its energy is
independent of that specific direction. This divergence between
the symmetry exhibited by the overall system and the symmetry
exhibited by the laws governing its evolution is an example of
spontaneous symmetry breaking. We have here a case where there
is a distinctively different law covering the N ->  system than
covers its individual constituents. This is exactly the kind of
difference of laws across levels of analysis that we noted
earlier as one criterion of a genuinely emergent phenomenon.
It is important to emphasize the qualitatively different
nature of these macroscopic phenomena, because it is not simply
a result of our inability to provide an explicit Schrödinger
equation for the macroscopic system (let alone solve it).
It is
a consequence of considering the macroscopic system as a system
composed of an infinite number of particles, with boundary
conditions that are macroscopic in form.
This means that the
system is necessarily nonmicroscopic, in that it is inconsistent
18
to consider a collection of infinitely many particles in
conceptually the same way as an atomistic component, which is
essentially singular.
These sorts of examples are persuasive, but they do of
course rely on idealized models of real systems.
That is in the
nature of the examples and should not count against them as
examples showing that emergent phenomena are not inescapably
mysterious, are perhaps common, occur within indisputably
physical systems rather than just biological or psychological
systems, and can be given a treatment that is far more detailed
than any supervenience account can even approach.
5. Conclusion
I believe that the arguments given here reveal three
things. First, that emergent properties are probably quite
common in the physical realm. If this is true, it will be likely
that there will be no sharp boundary between the physical level
and other levels. In turn, this will require refinement in terms
of what we mean by physicalism, and in our reasons for holding
it. Secondly, that the existence of these detailed models at the
physical level dissolve the air of mystery that has
traditionally surrounded emergentism and has led clear-headed
philosophers to stay away. Finally, that the level of detail
available in these models makes the use of supervenience
19
relations seem simplistic. And it is for these three reasons at
least that we need emergence, not supervenience.
REFERENCES
Anderson, P.W. (1972):"More is Different", Science 177: 393-396.
Armstrong, D.M. (1989): Universals: An Opinionated Introduction.
Boulder: Westview Press.
Humphreys, P.W. (1995), "Understanding in the Not-So-Special
Sciences", Southern Journal of Philosophy XXXIV, Supplement:
Spindel Conference 1995:99-114
_______________ (1997), 'How Properties Emerge', Philosophy of
Science 64: 53-70
20
____________
(1997a), "Aspects of Emergence", Philosophical
Topics 24: forthcoming.
Kim, J. (1993), Supervenience and Mind. Cambridge: Cambridge
University Press.
Lewis, D.K.(1986), On the Plurality of Worlds. Oxford:
Blackwell.
Rohrlich, F. (1997), `Cognitive Emergence', Philosophy of
Science 64 (Proceedings): This issue.
Rosenberg, A. (1997), `Can Physicalist Antireductionism Compute
the Embryo?', Philosophy of Science 64 (Proceedings): This
issue.
Sewell, G.L. (1986),
Quantum Theory of Collective Phenomena.
Oxford, The Clarendon Press.
Teller, P. (1992), `A Contemporary Look at Emergence', in A
Beckermann, H. Flohr, and J. Kim (eds), Emergence or Reduction?,
Berlin: Walter de Gruyter, 139-153.
Wimsatt, W. (1998): Emergence as Non-Aggregativity and the
Biases of Reductionisms", in W. Wimsatt, Piecewise
Approximations to Reality. Cambridge: Harvard University Press:
forthcoming.
21
FOOTNOTES
22
1I am here accepting only for expository purposes the idea that there exist distinct levels of
individuals and properties. Since most advocates of supervenience believe in this sort of layered
ontology, it would be a distraction to deny it here. So, for example, I here go along with the view
that atoms occupy a lower level than do simple inorganic molecules, which in turn occupy a
lower level than do complex organic molecules. These strata then provide a separation between
properties of atoms, simple inorganic molecules, and complex organic molecules. Arguments
against the levels view can be found in Humphreys (1995) and (1997a)
2.Interpretations of 'A supervenes on B' as 'A is nothing but B' can be found (in so many words)
in Armstrong (1989, 56):"...if it supervenes, I suggest, it is not distinct from what it supervenes
upon", and in Lewis (1986, 14):"...all there is to the picture is dots and non-dots at each point of
the matrix. The global properties are nothing but patterns in the dots."; "Symmetry is nothing but
a pattern in the arrangement of dots" (Lewis 1986, 15). See also Rosenberg (1997) for similar
sentiments about reduction.
3.Again, Armstrong (1989, 100):"The relation supervenes upon the terms...That, I think, makes
the relation an ontological free lunch."
4.In the discussion after the symposium, Andrew Melnyk suggested that some form of
necessitation other than nomological or metaphysical might well be appropriate for
supervenience relations. There was not enough time before submitting this paper to explore this
option.
5.This is the interpretation that seems to be favored by Kim throughout much of his (1993),
although he now seems to have doubts about the tenability of various supervenience views.
23
6This sort of case is canvassed, and rejected in Teller (1992).
7Whether it satisfies the third criterion depends upon how the levels are determined, something I
have put aside here.
8.Although I find Rohrlich's conception of emergence too psychological for my taste.
24
Download